1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * net/sched/act_police.c Input police filter 4 * 5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 6 * J Hadi Salim (action changes) 7 */ 8 9 #include <linux/module.h> 10 #include <linux/types.h> 11 #include <linux/kernel.h> 12 #include <linux/string.h> 13 #include <linux/errno.h> 14 #include <linux/skbuff.h> 15 #include <linux/rtnetlink.h> 16 #include <linux/init.h> 17 #include <linux/slab.h> 18 #include <net/act_api.h> 19 #include <net/netlink.h> 20 #include <net/pkt_cls.h> 21 #include <net/tc_act/tc_police.h> 22 23 /* Each policer is serialized by its individual spinlock */ 24 25 static unsigned int police_net_id; 26 static struct tc_action_ops act_police_ops; 27 28 static int tcf_police_walker(struct net *net, struct sk_buff *skb, 29 struct netlink_callback *cb, int type, 30 const struct tc_action_ops *ops, 31 struct netlink_ext_ack *extack) 32 { 33 struct tc_action_net *tn = net_generic(net, police_net_id); 34 35 return tcf_generic_walker(tn, skb, cb, type, ops, extack); 36 } 37 38 static const struct nla_policy police_policy[TCA_POLICE_MAX + 1] = { 39 [TCA_POLICE_RATE] = { .len = TC_RTAB_SIZE }, 40 [TCA_POLICE_PEAKRATE] = { .len = TC_RTAB_SIZE }, 41 [TCA_POLICE_AVRATE] = { .type = NLA_U32 }, 42 [TCA_POLICE_RESULT] = { .type = NLA_U32 }, 43 [TCA_POLICE_RATE64] = { .type = NLA_U64 }, 44 [TCA_POLICE_PEAKRATE64] = { .type = NLA_U64 }, 45 [TCA_POLICE_PKTRATE64] = { .type = NLA_U64, .min = 1 }, 46 [TCA_POLICE_PKTBURST64] = { .type = NLA_U64, .min = 1 }, 47 }; 48 49 static int tcf_police_init(struct net *net, struct nlattr *nla, 50 struct nlattr *est, struct tc_action **a, 51 struct tcf_proto *tp, u32 flags, 52 struct netlink_ext_ack *extack) 53 { 54 int ret = 0, tcfp_result = TC_ACT_OK, err, size; 55 bool bind = flags & TCA_ACT_FLAGS_BIND; 56 struct nlattr *tb[TCA_POLICE_MAX + 1]; 57 struct tcf_chain *goto_ch = NULL; 58 struct tc_police *parm; 59 struct tcf_police *police; 60 struct qdisc_rate_table *R_tab = NULL, *P_tab = NULL; 61 struct tc_action_net *tn = net_generic(net, police_net_id); 62 struct tcf_police_params *new; 63 bool exists = false; 64 u32 index; 65 u64 rate64, prate64; 66 u64 pps, ppsburst; 67 68 if (nla == NULL) 69 return -EINVAL; 70 71 err = nla_parse_nested_deprecated(tb, TCA_POLICE_MAX, nla, 72 police_policy, NULL); 73 if (err < 0) 74 return err; 75 76 if (tb[TCA_POLICE_TBF] == NULL) 77 return -EINVAL; 78 size = nla_len(tb[TCA_POLICE_TBF]); 79 if (size != sizeof(*parm) && size != sizeof(struct tc_police_compat)) 80 return -EINVAL; 81 82 parm = nla_data(tb[TCA_POLICE_TBF]); 83 index = parm->index; 84 err = tcf_idr_check_alloc(tn, &index, a, bind); 85 if (err < 0) 86 return err; 87 exists = err; 88 if (exists && bind) 89 return 0; 90 91 if (!exists) { 92 ret = tcf_idr_create(tn, index, NULL, a, 93 &act_police_ops, bind, true, flags); 94 if (ret) { 95 tcf_idr_cleanup(tn, index); 96 return ret; 97 } 98 ret = ACT_P_CREATED; 99 spin_lock_init(&(to_police(*a)->tcfp_lock)); 100 } else if (!(flags & TCA_ACT_FLAGS_REPLACE)) { 101 tcf_idr_release(*a, bind); 102 return -EEXIST; 103 } 104 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); 105 if (err < 0) 106 goto release_idr; 107 108 police = to_police(*a); 109 if (parm->rate.rate) { 110 err = -ENOMEM; 111 R_tab = qdisc_get_rtab(&parm->rate, tb[TCA_POLICE_RATE], NULL); 112 if (R_tab == NULL) 113 goto failure; 114 115 if (parm->peakrate.rate) { 116 P_tab = qdisc_get_rtab(&parm->peakrate, 117 tb[TCA_POLICE_PEAKRATE], NULL); 118 if (P_tab == NULL) 119 goto failure; 120 } 121 } 122 123 if (est) { 124 err = gen_replace_estimator(&police->tcf_bstats, 125 police->common.cpu_bstats, 126 &police->tcf_rate_est, 127 &police->tcf_lock, 128 false, est); 129 if (err) 130 goto failure; 131 } else if (tb[TCA_POLICE_AVRATE] && 132 (ret == ACT_P_CREATED || 133 !gen_estimator_active(&police->tcf_rate_est))) { 134 err = -EINVAL; 135 goto failure; 136 } 137 138 if (tb[TCA_POLICE_RESULT]) { 139 tcfp_result = nla_get_u32(tb[TCA_POLICE_RESULT]); 140 if (TC_ACT_EXT_CMP(tcfp_result, TC_ACT_GOTO_CHAIN)) { 141 NL_SET_ERR_MSG(extack, 142 "goto chain not allowed on fallback"); 143 err = -EINVAL; 144 goto failure; 145 } 146 } 147 148 if ((tb[TCA_POLICE_PKTRATE64] && !tb[TCA_POLICE_PKTBURST64]) || 149 (!tb[TCA_POLICE_PKTRATE64] && tb[TCA_POLICE_PKTBURST64])) { 150 NL_SET_ERR_MSG(extack, 151 "Both or neither packet-per-second burst and rate must be provided"); 152 err = -EINVAL; 153 goto failure; 154 } 155 156 if (tb[TCA_POLICE_PKTRATE64] && R_tab) { 157 NL_SET_ERR_MSG(extack, 158 "packet-per-second and byte-per-second rate limits not allowed in same action"); 159 err = -EINVAL; 160 goto failure; 161 } 162 163 new = kzalloc(sizeof(*new), GFP_KERNEL); 164 if (unlikely(!new)) { 165 err = -ENOMEM; 166 goto failure; 167 } 168 169 /* No failure allowed after this point */ 170 new->tcfp_result = tcfp_result; 171 new->tcfp_mtu = parm->mtu; 172 if (!new->tcfp_mtu) { 173 new->tcfp_mtu = ~0; 174 if (R_tab) 175 new->tcfp_mtu = 255 << R_tab->rate.cell_log; 176 } 177 if (R_tab) { 178 new->rate_present = true; 179 rate64 = tb[TCA_POLICE_RATE64] ? 180 nla_get_u64(tb[TCA_POLICE_RATE64]) : 0; 181 psched_ratecfg_precompute(&new->rate, &R_tab->rate, rate64); 182 qdisc_put_rtab(R_tab); 183 } else { 184 new->rate_present = false; 185 } 186 if (P_tab) { 187 new->peak_present = true; 188 prate64 = tb[TCA_POLICE_PEAKRATE64] ? 189 nla_get_u64(tb[TCA_POLICE_PEAKRATE64]) : 0; 190 psched_ratecfg_precompute(&new->peak, &P_tab->rate, prate64); 191 qdisc_put_rtab(P_tab); 192 } else { 193 new->peak_present = false; 194 } 195 196 new->tcfp_burst = PSCHED_TICKS2NS(parm->burst); 197 if (new->peak_present) 198 new->tcfp_mtu_ptoks = (s64)psched_l2t_ns(&new->peak, 199 new->tcfp_mtu); 200 201 if (tb[TCA_POLICE_AVRATE]) 202 new->tcfp_ewma_rate = nla_get_u32(tb[TCA_POLICE_AVRATE]); 203 204 if (tb[TCA_POLICE_PKTRATE64]) { 205 pps = nla_get_u64(tb[TCA_POLICE_PKTRATE64]); 206 ppsburst = nla_get_u64(tb[TCA_POLICE_PKTBURST64]); 207 new->pps_present = true; 208 new->tcfp_pkt_burst = PSCHED_TICKS2NS(ppsburst); 209 psched_ppscfg_precompute(&new->ppsrate, pps); 210 } 211 212 spin_lock_bh(&police->tcf_lock); 213 spin_lock_bh(&police->tcfp_lock); 214 police->tcfp_t_c = ktime_get_ns(); 215 police->tcfp_toks = new->tcfp_burst; 216 if (new->peak_present) 217 police->tcfp_ptoks = new->tcfp_mtu_ptoks; 218 spin_unlock_bh(&police->tcfp_lock); 219 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch); 220 new = rcu_replace_pointer(police->params, 221 new, 222 lockdep_is_held(&police->tcf_lock)); 223 spin_unlock_bh(&police->tcf_lock); 224 225 if (goto_ch) 226 tcf_chain_put_by_act(goto_ch); 227 if (new) 228 kfree_rcu(new, rcu); 229 230 return ret; 231 232 failure: 233 qdisc_put_rtab(P_tab); 234 qdisc_put_rtab(R_tab); 235 if (goto_ch) 236 tcf_chain_put_by_act(goto_ch); 237 release_idr: 238 tcf_idr_release(*a, bind); 239 return err; 240 } 241 242 static bool tcf_police_mtu_check(struct sk_buff *skb, u32 limit) 243 { 244 u32 len; 245 246 if (skb_is_gso(skb)) 247 return skb_gso_validate_mac_len(skb, limit); 248 249 len = qdisc_pkt_len(skb); 250 if (skb_at_tc_ingress(skb)) 251 len += skb->mac_len; 252 253 return len <= limit; 254 } 255 256 static int tcf_police_act(struct sk_buff *skb, const struct tc_action *a, 257 struct tcf_result *res) 258 { 259 struct tcf_police *police = to_police(a); 260 s64 now, toks, ppstoks = 0, ptoks = 0; 261 struct tcf_police_params *p; 262 int ret; 263 264 tcf_lastuse_update(&police->tcf_tm); 265 bstats_update(this_cpu_ptr(police->common.cpu_bstats), skb); 266 267 ret = READ_ONCE(police->tcf_action); 268 p = rcu_dereference_bh(police->params); 269 270 if (p->tcfp_ewma_rate) { 271 struct gnet_stats_rate_est64 sample; 272 273 if (!gen_estimator_read(&police->tcf_rate_est, &sample) || 274 sample.bps >= p->tcfp_ewma_rate) 275 goto inc_overlimits; 276 } 277 278 if (tcf_police_mtu_check(skb, p->tcfp_mtu)) { 279 if (!p->rate_present && !p->pps_present) { 280 ret = p->tcfp_result; 281 goto end; 282 } 283 284 now = ktime_get_ns(); 285 spin_lock_bh(&police->tcfp_lock); 286 toks = min_t(s64, now - police->tcfp_t_c, p->tcfp_burst); 287 if (p->peak_present) { 288 ptoks = toks + police->tcfp_ptoks; 289 if (ptoks > p->tcfp_mtu_ptoks) 290 ptoks = p->tcfp_mtu_ptoks; 291 ptoks -= (s64)psched_l2t_ns(&p->peak, 292 qdisc_pkt_len(skb)); 293 } 294 if (p->rate_present) { 295 toks += police->tcfp_toks; 296 if (toks > p->tcfp_burst) 297 toks = p->tcfp_burst; 298 toks -= (s64)psched_l2t_ns(&p->rate, qdisc_pkt_len(skb)); 299 } else if (p->pps_present) { 300 ppstoks = min_t(s64, now - police->tcfp_t_c, p->tcfp_pkt_burst); 301 ppstoks += police->tcfp_pkttoks; 302 if (ppstoks > p->tcfp_pkt_burst) 303 ppstoks = p->tcfp_pkt_burst; 304 ppstoks -= (s64)psched_pkt2t_ns(&p->ppsrate, 1); 305 } 306 if ((toks | ptoks | ppstoks) >= 0) { 307 police->tcfp_t_c = now; 308 police->tcfp_toks = toks; 309 police->tcfp_ptoks = ptoks; 310 police->tcfp_pkttoks = ppstoks; 311 spin_unlock_bh(&police->tcfp_lock); 312 ret = p->tcfp_result; 313 goto inc_drops; 314 } 315 spin_unlock_bh(&police->tcfp_lock); 316 } 317 318 inc_overlimits: 319 qstats_overlimit_inc(this_cpu_ptr(police->common.cpu_qstats)); 320 inc_drops: 321 if (ret == TC_ACT_SHOT) 322 qstats_drop_inc(this_cpu_ptr(police->common.cpu_qstats)); 323 end: 324 return ret; 325 } 326 327 static void tcf_police_cleanup(struct tc_action *a) 328 { 329 struct tcf_police *police = to_police(a); 330 struct tcf_police_params *p; 331 332 p = rcu_dereference_protected(police->params, 1); 333 if (p) 334 kfree_rcu(p, rcu); 335 } 336 337 static void tcf_police_stats_update(struct tc_action *a, 338 u64 bytes, u64 packets, u64 drops, 339 u64 lastuse, bool hw) 340 { 341 struct tcf_police *police = to_police(a); 342 struct tcf_t *tm = &police->tcf_tm; 343 344 tcf_action_update_stats(a, bytes, packets, drops, hw); 345 tm->lastuse = max_t(u64, tm->lastuse, lastuse); 346 } 347 348 static int tcf_police_dump(struct sk_buff *skb, struct tc_action *a, 349 int bind, int ref) 350 { 351 unsigned char *b = skb_tail_pointer(skb); 352 struct tcf_police *police = to_police(a); 353 struct tcf_police_params *p; 354 struct tc_police opt = { 355 .index = police->tcf_index, 356 .refcnt = refcount_read(&police->tcf_refcnt) - ref, 357 .bindcnt = atomic_read(&police->tcf_bindcnt) - bind, 358 }; 359 struct tcf_t t; 360 361 spin_lock_bh(&police->tcf_lock); 362 opt.action = police->tcf_action; 363 p = rcu_dereference_protected(police->params, 364 lockdep_is_held(&police->tcf_lock)); 365 opt.mtu = p->tcfp_mtu; 366 opt.burst = PSCHED_NS2TICKS(p->tcfp_burst); 367 if (p->rate_present) { 368 psched_ratecfg_getrate(&opt.rate, &p->rate); 369 if ((police->params->rate.rate_bytes_ps >= (1ULL << 32)) && 370 nla_put_u64_64bit(skb, TCA_POLICE_RATE64, 371 police->params->rate.rate_bytes_ps, 372 TCA_POLICE_PAD)) 373 goto nla_put_failure; 374 } 375 if (p->peak_present) { 376 psched_ratecfg_getrate(&opt.peakrate, &p->peak); 377 if ((police->params->peak.rate_bytes_ps >= (1ULL << 32)) && 378 nla_put_u64_64bit(skb, TCA_POLICE_PEAKRATE64, 379 police->params->peak.rate_bytes_ps, 380 TCA_POLICE_PAD)) 381 goto nla_put_failure; 382 } 383 if (p->pps_present) { 384 if (nla_put_u64_64bit(skb, TCA_POLICE_PKTRATE64, 385 police->params->ppsrate.rate_pkts_ps, 386 TCA_POLICE_PAD)) 387 goto nla_put_failure; 388 if (nla_put_u64_64bit(skb, TCA_POLICE_PKTBURST64, 389 PSCHED_NS2TICKS(p->tcfp_pkt_burst), 390 TCA_POLICE_PAD)) 391 goto nla_put_failure; 392 } 393 if (nla_put(skb, TCA_POLICE_TBF, sizeof(opt), &opt)) 394 goto nla_put_failure; 395 if (p->tcfp_result && 396 nla_put_u32(skb, TCA_POLICE_RESULT, p->tcfp_result)) 397 goto nla_put_failure; 398 if (p->tcfp_ewma_rate && 399 nla_put_u32(skb, TCA_POLICE_AVRATE, p->tcfp_ewma_rate)) 400 goto nla_put_failure; 401 402 tcf_tm_dump(&t, &police->tcf_tm); 403 if (nla_put_64bit(skb, TCA_POLICE_TM, sizeof(t), &t, TCA_POLICE_PAD)) 404 goto nla_put_failure; 405 spin_unlock_bh(&police->tcf_lock); 406 407 return skb->len; 408 409 nla_put_failure: 410 spin_unlock_bh(&police->tcf_lock); 411 nlmsg_trim(skb, b); 412 return -1; 413 } 414 415 static int tcf_police_search(struct net *net, struct tc_action **a, u32 index) 416 { 417 struct tc_action_net *tn = net_generic(net, police_net_id); 418 419 return tcf_idr_search(tn, a, index); 420 } 421 422 static int tcf_police_act_to_flow_act(int tc_act, u32 *extval) 423 { 424 int act_id = -EOPNOTSUPP; 425 426 if (!TC_ACT_EXT_OPCODE(tc_act)) { 427 if (tc_act == TC_ACT_OK) 428 act_id = FLOW_ACTION_ACCEPT; 429 else if (tc_act == TC_ACT_SHOT) 430 act_id = FLOW_ACTION_DROP; 431 else if (tc_act == TC_ACT_PIPE) 432 act_id = FLOW_ACTION_PIPE; 433 } else if (TC_ACT_EXT_CMP(tc_act, TC_ACT_GOTO_CHAIN)) { 434 act_id = FLOW_ACTION_GOTO; 435 *extval = tc_act & TC_ACT_EXT_VAL_MASK; 436 } else if (TC_ACT_EXT_CMP(tc_act, TC_ACT_JUMP)) { 437 act_id = FLOW_ACTION_JUMP; 438 *extval = tc_act & TC_ACT_EXT_VAL_MASK; 439 } 440 441 return act_id; 442 } 443 444 static int tcf_police_offload_act_setup(struct tc_action *act, void *entry_data, 445 u32 *index_inc, bool bind) 446 { 447 if (bind) { 448 struct flow_action_entry *entry = entry_data; 449 struct tcf_police *police = to_police(act); 450 struct tcf_police_params *p; 451 int act_id; 452 453 p = rcu_dereference_protected(police->params, 454 lockdep_is_held(&police->tcf_lock)); 455 456 entry->id = FLOW_ACTION_POLICE; 457 entry->police.burst = tcf_police_burst(act); 458 entry->police.rate_bytes_ps = 459 tcf_police_rate_bytes_ps(act); 460 entry->police.peakrate_bytes_ps = tcf_police_peakrate_bytes_ps(act); 461 entry->police.avrate = tcf_police_tcfp_ewma_rate(act); 462 entry->police.overhead = tcf_police_rate_overhead(act); 463 entry->police.burst_pkt = tcf_police_burst_pkt(act); 464 entry->police.rate_pkt_ps = 465 tcf_police_rate_pkt_ps(act); 466 entry->police.mtu = tcf_police_tcfp_mtu(act); 467 468 act_id = tcf_police_act_to_flow_act(police->tcf_action, 469 &entry->police.exceed.extval); 470 if (act_id < 0) 471 return act_id; 472 473 entry->police.exceed.act_id = act_id; 474 475 act_id = tcf_police_act_to_flow_act(p->tcfp_result, 476 &entry->police.notexceed.extval); 477 if (act_id < 0) 478 return act_id; 479 480 entry->police.notexceed.act_id = act_id; 481 482 *index_inc = 1; 483 } else { 484 struct flow_offload_action *fl_action = entry_data; 485 486 fl_action->id = FLOW_ACTION_POLICE; 487 } 488 489 return 0; 490 } 491 492 MODULE_AUTHOR("Alexey Kuznetsov"); 493 MODULE_DESCRIPTION("Policing actions"); 494 MODULE_LICENSE("GPL"); 495 496 static struct tc_action_ops act_police_ops = { 497 .kind = "police", 498 .id = TCA_ID_POLICE, 499 .owner = THIS_MODULE, 500 .stats_update = tcf_police_stats_update, 501 .act = tcf_police_act, 502 .dump = tcf_police_dump, 503 .init = tcf_police_init, 504 .walk = tcf_police_walker, 505 .lookup = tcf_police_search, 506 .cleanup = tcf_police_cleanup, 507 .offload_act_setup = tcf_police_offload_act_setup, 508 .size = sizeof(struct tcf_police), 509 }; 510 511 static __net_init int police_init_net(struct net *net) 512 { 513 struct tc_action_net *tn = net_generic(net, police_net_id); 514 515 return tc_action_net_init(net, tn, &act_police_ops); 516 } 517 518 static void __net_exit police_exit_net(struct list_head *net_list) 519 { 520 tc_action_net_exit(net_list, police_net_id); 521 } 522 523 static struct pernet_operations police_net_ops = { 524 .init = police_init_net, 525 .exit_batch = police_exit_net, 526 .id = &police_net_id, 527 .size = sizeof(struct tc_action_net), 528 }; 529 530 static int __init police_init_module(void) 531 { 532 return tcf_register_action(&act_police_ops, &police_net_ops); 533 } 534 535 static void __exit police_cleanup_module(void) 536 { 537 tcf_unregister_action(&act_police_ops, &police_net_ops); 538 } 539 540 module_init(police_init_module); 541 module_exit(police_cleanup_module); 542